On-device machine learning (ML) moves cloud computing to personal devices, protecting user privacy and enabling intelligent user experiences. However, tailoring models to resource-constrained devices presents a significant technical challenge: practitioners must optimize models and balance hardware metrics such as model size, latency, and power. To help professionals create efficient machine learning models, we designed and developed Talaria: a model visualization and optimization system. Talaria allows professionals to build models on hardware, interactively visualize model statistics, and simulate optimizations to test the impact on inference metrics. Since its internal implementation two years ago, we have evaluated Talaria using three methodologies: (1) a log analysis highlighting its growth of over 800 professionals submitting over 3,600 models; (2) a usability survey with 26 users evaluating the usefulness of 20 Talaria features; and (3) a qualitative interview with the 7 most active users about their experience using Talaria.
